Skip to content
Cerberus

Defensive work, for the people who own the software

We protect other people’s software for a living, so this page says how we handle it, how we do our research, and where we draw the line.

Research

Our protection is only as good as our understanding of the attack, and the attack is now a language model. So part of our work is offensive in method and defensive in purpose: we point current models at software we protect, see what they recover, and harden the pipeline against it.

That research runs only against our own builds and builds whose owners have asked us to test them. It follows the usage policies of the model providers we work with. What we learn goes into protection, not into public write-ups that would teach the same models where to look.

Model providers and security researchers who want to know more can write to us.

Responsible use

Software protection is dual-use. The techniques that keep a paid product from being cracked could also hide code that harms people. We only work with teams protecting software they own or are licensed to protect.

  • You need the rights to anything you send us to protect.
  • We do not protect malware, spyware or code meant to evade security products.
  • We do not break other people’s protection.
  • Watermarks trace leaks back to a license. They are not for watching users.

We decline work that falls outside this, and withdraw access if use changes.

Handling

Source code
Never needs to reach us. We work on built artifacts.
Keys and tooling
Stay on our side. Nothing secret ships inside a protected build.
Isolation
Every protection job runs in its own environment. No job can see another.
Retention
Inputs and outputs live as long as a release channel needs them, then age out.
Audit
Pipeline runs, issued builds and staff actions are logged.
Certification
Not yet audited against SOC 2 or ISO 27001. Controls are built toward both.

Disclosure

Found a weakness in the platform or in a build it produced? Tell us first and give us a reasonable window to fix it. We will not pursue good-faith research that respects user privacy and avoids disruption. [email protected]